MOTOMIND
TECHNOLOGY

We build the body, teach the motion, and share the intelligence

MOTOMIND develops the actuators and robot structure, the edge board, the learning pipeline, and the shared runtime in one team. A human demonstration becomes a robot task, and a task one robot has learned carries over to other robots and sites — all on a single stack.

30 DoF
BELLO whole-body degrees of freedom
20 kg
BELLO whole-body payload
10 kg
BELLO mini single-arm payload
275 TOPS
AI Driver peak compute (mmRA-20)
01ROBOT HARDWARE

Low-reduction, high-torque actuators and an endoskeleton structure

Pancake-motor actuators with low gear reduction keep reflected inertia low and preserve backdrivability. Arms are built as an endoskeleton rather than an exoskeleton, cutting volume and cost, and the same actuator modules are reused from the MILLY arm to the BELLO and BELLO mini humanoids.

Actuator
Pancake BLDC · low reduction, high torque
Structure
Endoskeleton arm · shared modules throughout
Humanoids
BELLO 170 cm · BELLO mini 140 cm, 30 DoF
Arm
MILLY 6+1 DoF · 3 kg payload
Locomotion
RL-based bipedal walking (NextRise 2026 demo)
See the product line-up
BELLO — unsupported bipedal walking, NextRise 2026
BELLO mini — single-arm 10 kg kettlebell lift
MILLY — endoskeleton 6+1 DoF AI robot arm
MILLY — endoskeleton 6+1 DoF AI robot arm
02AI DRIVER

Control, sensing, and compute on one edge board

A carrier board for NVIDIA Jetson modules that integrates motor drivers, an IMU and barometer, an STM32 real-time MCU, and CAN / RS-485. Arms, humanoids, and mobile platforms all run on the same board and firmware, so a control stack validated on one robot moves to the next unchanged.

Compute
Jetson Orin Nano / NX 40–100 TOPS · AGX Orin 275 TOPS
Drive
2-ch motor driver · up to 250 W per channel (mmRA-10)
Real-time
STM32G474 MCU · CAN FD · RS-232 / RS-485
Sensors
IMU LSM6DSR · barometer BMP388 · power monitoring
Expansion
M.2 KEY-M AI accelerator · NVMe · M.2 KEY-E WiFi / BT
AI Board details
mmRA-10 — top-side interface map
mmRA-10 — top-side interface map
mmRA-20 — Jetson AGX Orin carrier board
mmRA-20 — Jetson AGX Orin carrier board
mmRA-10 — assembled board (MOTOMIND v1.2)
mmRA-10 — assembled board (MOTOMIND v1.2)
LEARNING

From human motion to robot work

Collect operator demonstrations as data, map them onto the robot's body, verify in simulation, then execute on the real robot. Data gathered in operation flows back into training.

  1. 01

    Teach — demonstrations become data

    Operators drive the robot directly through teleoperation to demonstrate a task. Haptic assist corrects for skill variance so the demonstration data stays consistent.

    MILLY teleoperation — collecting task demonstrations
  2. 02

    Retarget — fitted to the robot's body

    Whole-body motion captured with VR and motion tracking is mapped in real time onto the robot's joint structure. The retargeted motion is used directly as training data.

    BELLO mini — real-time whole-body retargeting
  3. 03

    Simulate — verified in the virtual world first

    Motions are verified in a simulator that models the real response of our own actuators. Real tasks are reproduced virtually (real-to-sim) and policies are refined with reinforcement learning.

    Real-to-sim reproduction of the box-folding task
  4. 04

    Execute — real work on the real robot

    Verified policies run on the AI Driver. Camera perception and the learned policy perform the task, and operating data flows back into training.

    Dual-arm MILLY — stacking dishes autonomously with a trained VLA policy
RUNTIME · SDK

Many robots, one shared intelligence

Instead of a GPU and model per robot, a shared AI Runtime handles inference, planning, and model versioning while each robot's AI Driver focuses on sensing, real-time control, and safety stops. A common SDK joins the two layers.

  1. EDGE

    Robots · AI Driver

    • Sensing · real-time control
    • Safety stop · power monitoring
    • MILLY · BELLO · BELLO mini
  2. SHARED

    Physical AI Runtime

    • Shared reasoning · action inference
    • Stateful sessions · RTC streaming
    • Model versioning · fleet-wide rollout
  3. SDK

    Common SDK · API

    • Python SDK · ROS2 · common API
    • Run one robot's task on another
    • Connects to existing robotics stacks

The Runtime is deployed on-premises or in the cloud depending on the customer's security requirements.

Let's build real-world work together

We welcome hardware adoption, demo, and joint-development / PoC inquiries.

Talk to usSee products